The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external translation strings sourced from Crowdin projects. Maliciously crafted strings within these projects could potentially influence the AI's generated output for the context fields.
Ingestion points: The crowdin-context.jsonl file, which is downloaded directly from the Crowdin platform.
Boundary markers: The instructions explicitly restrict the agent to editing only the ai_context field and provide a detailed validity checklist to ensure JSON integrity.
Capability inventory: The skill performs local file reading (JSONL and source code), file writing (JSONL), and shell command execution via the crowdin CLI for data synchronization.
Sanitization: Clear guidelines are provided for JSON character escaping and structure verification before saving changes.
[COMMAND_EXECUTION]: The skill utilizes the crowdin CLI tool to perform several operations, including checking status, downloading translation context, and uploading modified data. These commands represent the core functionality of the skill and involve network communication with the Crowdin service.

context-extraction